Accelerating Convolutional Neural Networks using General Purpose Processing on Graphical Processing Unit
Z. Ciota · 2009
General purpose processing is relatively new approach, which has recently become much more widely adopted, thanks to GPU manufacturers providing programming solutions and very recently thanks to OpenCL standard provided by Khronos group. CUDA used in the experiment described in this article is an extension of C programming language, enables programming heterogeneous systems. Convolutional Neural Networks are biologically inspired. They are based on the concept of “simple” and “complex” neuron cells[9]. CNNs are designed with multiple layers of neurons, use weight sharing to reduce the number of free parameters[3]. They exhibit convolution which occurs between layers[4]